A Game-Theoretic Approach to Personnel Decisions in American ...
A Game-Theoretic Approach to Personnel Decisions in American ... A Game-Theoretic Approach to Personnel Decisions in American ...
Chicago Bears this past season and some implications due to this comparison. 2 Setting Up the Game We consider two-by-two zero-sum games where the offense either runs or passes and the defense either defends against the run or defends against the pass. Each payoff represents an expected gain for the offense under the given strategic situation. For any serious football team, it is reasonable to suggest that the payoffs can be found, since every coaching staff spends much of their time breaking down film. Table 1: Initial Payoffs Defense Defend Run Defend Pass Offense Run ar,r ar,p Pass ap,r ap,p Let the offensive and defensive strategy sets be {R,P } and {DR,DP }, respectively. Furthermore, let ar,r denote the payoff for (R,DR), ar,p denote the payoff for (R,DP), and so on, as in Table 1. Throughout this paper, we make the reasonable assumptions that ar,r < ap,r and ar,p > ap,p. We calculate the expected gain for the offensive team in a straightforward manner, save that we adopt a convention used by Alamar (2006) and Rockerbie (2008) and deduct 45 yards from a team’s total yardage for each fumble and interception. For instance, to determine ar,r, suppose that there were nr,r observed plays in which the offense elected to run and the defense elected to defend the run, amounting in a total of yr,r yards and tr,r turnovers. Then ar,r = yr,r − 45tr,r . The remaining payoffs in Table 1 are calculated in a similar manner. nr,r 3 The Improvement in Passing Model Having demonstrated how to calculate the payoffs for our team prior to the relevant personnel change, we now model the impact of this personnel change
on our team’s performance. We assume that the team has acquired a proven quarterback whose presence will improve all aspects of the passing game. This will be reflected in our payoff table via an increase in both payoff entries associated with an offensive passing play. The following additional assumption is a critical part of our model. It is a common belief that the success of the passing game is intertwined with the success of the running game. Consequently, we suppose that the increase in expected gains through passing will also have a positive effect on the running payoffs ar,r and ar,p. Although the quarterback will positively influence the running game, he will have a greater impact on the passing game. This is because production in the passing game rests squarely upon his ability and performance, whereas the quarterback’s influence is generally limited in the running game. We now introduce the model. Throughout this analysis, we assume that the payoff matrix depicted in Table 1 contains no dominant rows and no dominant columns, implying that our games will have only mixed strategy Nash equilibria (Nash (1951)). This assumption is reasonable in the context of our desired applications, as it is unlikely that a serious amateur or professional football team would ever utilize a pure offensive strategy. We let x represent the proportional increase in the passing game for some real x > 0 and let δ measure the residual effect of the improved passing game on the running game. We conceptualize x as a (relative) quarterback rating, so that if x = 0, the new quarterback is no better than the previous one. These parameters are incorporated into the payoffs from Table 1 to obtain the 2-by-2 game in 2, which reflects the assumed changes in player personnel. We refer to this as the improvement in passing model . Defense Defend Run Defend Pass Offense Run ar,r(1 + δx) ar,p(1 + δx) Pass ap,r(1 + x) ap,p(1 + x) Table 2: The Improvement in Passing Model Comparing mobile quarterbacks and pocket passers of the same caliber, it is reasonable to suggest that mobile quarterbacks have a higher δ value, since they are contributing to both phases of the game. In the case of a highly immobile quarterback with a strong arm, the value of δ for each quarterback could, in fact, be negative, even if x is relatively high. While this is feasible,
- Page 1: 1 Introduction There is a growing i
- Page 5 and 6: As a notational convenience, let D(
- Page 7 and 8: ∂R∗ ∂x = (ar,p − ar,r)(ap,r
- Page 9 and 10: in order to produce our desired pay
- Page 11 and 12: Therefore, if we assume that Jay Cu
- Page 13: Carter, V. and R. Machol (1978):
on our team’s performance. We assume that the team has acquired a proven<br />
quarterback whose presence will improve all aspects of the pass<strong>in</strong>g game. This<br />
will be reflected <strong>in</strong> our payoff table via an <strong>in</strong>crease <strong>in</strong> both payoff entries<br />
associated with an offensive pass<strong>in</strong>g play.<br />
The follow<strong>in</strong>g additional assumption is a critical part of our model. It<br />
is a common belief that the success of the pass<strong>in</strong>g game is <strong>in</strong>tertw<strong>in</strong>ed with the<br />
success of the runn<strong>in</strong>g game. Consequently, we suppose that the <strong>in</strong>crease <strong>in</strong><br />
expected ga<strong>in</strong>s through pass<strong>in</strong>g will also have a positive effect on the runn<strong>in</strong>g<br />
payoffs ar,r and ar,p. Although the quarterback will positively <strong>in</strong>fluence the<br />
runn<strong>in</strong>g game, he will have a greater impact on the pass<strong>in</strong>g game. This is<br />
because production <strong>in</strong> the pass<strong>in</strong>g game rests squarely upon his ability and<br />
performance, whereas the quarterback’s <strong>in</strong>fluence is generally limited <strong>in</strong> the<br />
runn<strong>in</strong>g game. We now <strong>in</strong>troduce the model.<br />
Throughout this analysis, we assume that the payoff matrix depicted<br />
<strong>in</strong> Table 1 conta<strong>in</strong>s no dom<strong>in</strong>ant rows and no dom<strong>in</strong>ant columns, imply<strong>in</strong>g<br />
that our games will have only mixed strategy Nash equilibria (Nash (1951)).<br />
This assumption is reasonable <strong>in</strong> the context of our desired applications, as<br />
it is unlikely that a serious amateur or professional football team would ever<br />
utilize a pure offensive strategy.<br />
We let x represent the proportional <strong>in</strong>crease <strong>in</strong> the pass<strong>in</strong>g game for<br />
some real x > 0 and let δ measure the residual effect of the improved pass<strong>in</strong>g<br />
game on the runn<strong>in</strong>g game. We conceptualize x as a (relative) quarterback<br />
rat<strong>in</strong>g, so that if x = 0, the new quarterback is no better than the previous one.<br />
These parameters are <strong>in</strong>corporated <strong>in</strong><strong>to</strong> the payoffs from Table 1 <strong>to</strong> obta<strong>in</strong> the<br />
2-by-2 game <strong>in</strong> 2, which reflects the assumed changes <strong>in</strong> player personnel. We<br />
refer <strong>to</strong> this as the improvement <strong>in</strong> pass<strong>in</strong>g model .<br />
Defense<br />
Defend Run Defend Pass<br />
Offense Run ar,r(1 + δx) ar,p(1 + δx)<br />
Pass ap,r(1 + x) ap,p(1 + x)<br />
Table 2: The Improvement <strong>in</strong> Pass<strong>in</strong>g Model<br />
Compar<strong>in</strong>g mobile quarterbacks and pocket passers of the same caliber,<br />
it is reasonable <strong>to</strong> suggest that mobile quarterbacks have a higher δ value,<br />
s<strong>in</strong>ce they are contribut<strong>in</strong>g <strong>to</strong> both phases of the game. In the case of a highly<br />
immobile quarterback with a strong arm, the value of δ for each quarterback<br />
could, <strong>in</strong> fact, be negative, even if x is relatively high. While this is feasible,
Change language